| 摘要: |
| 传统的数据相关性分析主要针对线性数组下标表达式,并不适用于不规则计算中的循环并行性识别.利用间接数组特征分析和基于严格数组私有化定义的运行时动态测试技术来解决包含间接数组下标引用的循环并行性判断问题,给出性能估测,并与相关工作进行了比较. |
| 关键词: 并行识别 不规则 数组私有化 运行时测试 |
| DOI： |
| 分类号: |
| 基金项目:国防科技重点实验室基金资助项目(JS94.6.1JW0703);国家教育部博士点基金资助项目;上海市青年科技启明星计划资助项目(99QD14043) |
|
| Automatic Parallelism Detection for One Kind of Irregular Problems |
|
LI Jing,ZANG Bin-yu,ZHU Chuan-qi
|
| Abstract: |
| Traditional data dependence analysis focuses on affine subscript, which is not applicable to detect parallelism in irregular problem. In this paper, two analysis techniques for subscripted subscripts are presented. One takes the property of indirect array into accounts, the other uses runtime test based on the strict array privatization definition. Comparison between the existing methods and the new techniques is also given. |
| Key words: parallelism detection irregular array privatization runtime test |
